Page MenuHomePhabricator
Feed Advanced Search

Yesterday

rampitec committed rGb37d6a750a0f: [AMDGPU] Check for immediate SrcC in mfma in AsmParser (authored by rampitec).
[AMDGPU] Check for immediate SrcC in mfma in AsmParser
Fri, Aug 23, 3:24 PM
rampitec committed rGe6e1c4eac042: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug (authored by rampitec).
[AMDGPU] w/a for gfx908 mfma SrcC literal HW bug
Fri, Aug 23, 3:23 PM
rampitec committed rL369819: [AMDGPU] Check for immediate SrcC in mfma in AsmParser.
[AMDGPU] Check for immediate SrcC in mfma in AsmParser
Fri, Aug 23, 3:23 PM
rampitec closed D66674: [AMDGPU] Check for immediate SrcC in mfma in AsmParser.
Fri, Aug 23, 3:23 PM · Restricted Project
rampitec committed rL369818: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.
[AMDGPU] w/a for gfx908 mfma SrcC literal HW bug
Fri, Aug 23, 3:23 PM
rampitec committed rG8fe1245a0f75: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug (authored by rampitec).
[AMDGPU] w/a for gfx908 mfma SrcC literal HW bug
Fri, Aug 23, 3:14 PM
rampitec committed rL369816: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.
[AMDGPU] w/a for gfx908 mfma SrcC literal HW bug
Fri, Aug 23, 3:14 PM
rampitec closed D66670: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.
Fri, Aug 23, 3:14 PM · Restricted Project
rampitec updated the diff for D66670: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.

Moved check into opCanUseInlineConstant().
Captured subtarget in SIRegisterInfo to do this.

Fri, Aug 23, 2:32 PM · Restricted Project
rampitec added a comment to D66670: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.

Can these be encoded as a constant literal instead? We would then just need to workaround the encoding rather than disallowing the operand

Fri, Aug 23, 1:15 PM · Restricted Project
rampitec added a parent revision for D66674: [AMDGPU] Check for immediate SrcC in mfma in AsmParser: D66670: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.
Fri, Aug 23, 12:38 PM · Restricted Project
rampitec added a child revision for D66670: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug: D66674: [AMDGPU] Check for immediate SrcC in mfma in AsmParser.
Fri, Aug 23, 12:38 PM · Restricted Project
rampitec created D66674: [AMDGPU] Check for immediate SrcC in mfma in AsmParser.
Fri, Aug 23, 12:38 PM · Restricted Project
rampitec added inline comments to D66670: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.
Fri, Aug 23, 12:18 PM · Restricted Project
rampitec added a comment to D66670: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.

Can we have different opcodes with a different source type?

It will be a quite big change. What for?

Generally if the operand/register class changes there should be a separate opcode. It avoids needing all this special case checking, and makes folding code work naturally

Anyway, you cannot have two opcodes with the same opcode without introducing a new encoding and a new decoder namespace. This bug will be fixed.

The encoding is the same, and a new decoder namespace shouldn't be an issue

Fri, Aug 23, 12:18 PM · Restricted Project
rampitec added a comment to D66670: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.

Can we have different opcodes with a different source type?

It will be a quite big change. What for?

Generally if the operand/register class changes there should be a separate opcode. It avoids needing all this special case checking, and makes folding code work naturally

Fri, Aug 23, 12:01 PM · Restricted Project
rampitec added a comment to D66670: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.

Can we have different opcodes with a different source type?

Fri, Aug 23, 11:48 AM · Restricted Project
rampitec created D66670: [AMDGPU] w/a for gfx908 mfma SrcC literal HW bug.
Fri, Aug 23, 11:39 AM · Restricted Project

Thu, Aug 22

rampitec accepted D66629: AMDGPU: Don't use frame virtual registers.

LGTM

Thu, Aug 22, 8:01 PM
rampitec added inline comments to D62739: AMDGPU: Always emit amdgpu-flat-work-group-size.
Thu, Aug 22, 12:08 PM

Tue, Aug 20

rampitec accepted D64335: AMDGPU: Move SelectFlatOffset back into AMDGPUISelDAGToDAG.

LGTM

Tue, Aug 20, 4:58 PM

Mon, Aug 19

rampitec added a comment to D66430: [AMDGPU] Kernel calls lowering.

This pass should be removed. Clang needs to emit a separate function body up-front. I want the IR verifier to reject all calls to kernels. Since a module may need to export a callable function with the name of the kernel, this has a name mangling impact that the frontend needs to take care of

I agree in general. However, we do not support late linking so far but do support OpenCL. Is there any reason to keep this pass even as a temporary solution in the private branch?

I can't imagine the clang patch is a lot of work, but nobody's ever done it. Ignoring the name mangling problem, it shouldn't be difficult to just emit the two separate functions directly

Mon, Aug 19, 11:11 AM
rampitec added a comment to D66430: [AMDGPU] Kernel calls lowering.

This pass should be removed. Clang needs to emit a separate function body up-front. I want the IR verifier to reject all calls to kernels. Since a module may need to export a callable function with the name of the kernel, this has a name mangling impact that the frontend needs to take care of

Mon, Aug 19, 11:00 AM
rampitec created D66430: [AMDGPU] Kernel calls lowering.
Mon, Aug 19, 10:47 AM

Sat, Aug 17

rampitec accepted D66382: AMDGPU: Fix iterator error when lowering SI_END_CF.

LGTM

Sat, Aug 17, 2:07 PM

Wed, Aug 14

rampitec committed rGc17705b7fb19: [AMDGPU] Do not assume a default GCN target (authored by rampitec).
[AMDGPU] Do not assume a default GCN target
Wed, Aug 14, 1:59 PM
rampitec committed rL368917: [AMDGPU] Do not assume a default GCN target.
[AMDGPU] Do not assume a default GCN target
Wed, Aug 14, 1:58 PM
rampitec closed D66246: [AMDGPU] Do not assume a default GCN target.
Wed, Aug 14, 1:58 PM · Restricted Project
rampitec added inline comments to D66246: [AMDGPU] Do not assume a default GCN target.
Wed, Aug 14, 1:30 PM · Restricted Project
rampitec created D66246: [AMDGPU] Do not assume a default GCN target.
Wed, Aug 14, 1:02 PM · Restricted Project
rampitec added a comment to D66198: AMDGPU: Add builtins for is_local/is_private.

Didn't you forget to update test/CodeGenOpenCL/amdgpu-features.cl?

Wed, Aug 14, 11:19 AM
rampitec accepted D66198: AMDGPU: Add builtins for is_local/is_private.

LGTM

Wed, Aug 14, 11:05 AM
rampitec accepted D66197: AMDGPU: Add intrinsics for address space identification.

LGTM

Wed, Aug 14, 11:05 AM

Mon, Aug 12

rampitec committed rG438315bf69c7: [AMDGPU] Fix msan failure in printf lowering (authored by rampitec).
[AMDGPU] Fix msan failure in printf lowering
Mon, Aug 12, 6:11 PM
rampitec added a comment to D24035: [AMDGPU] Printf runtime binding pass.

It's a bit hard to use, you need to build libcxx/libcxxabi with msan as well.
See here:
https://github.com/google/sanitizers/wiki/MemorySanitizerBootstrappingClang

I've started a build with track-origins here:
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-bootstrap-msan/builds/14114
It could give more information.

Mon, Aug 12, 6:10 PM · Restricted Project
rampitec committed rL368645: [AMDGPU] Fix msan failure in printf lowering.
[AMDGPU] Fix msan failure in printf lowering
Mon, Aug 12, 6:06 PM
rampitec added a comment to D24035: [AMDGPU] Printf runtime binding pass.

Hi, MSan is complaining about this change:
==28574==WARNING: MemorySanitizer: use-of-uninitialized-value

#0 0x130075d in (anonymous namespace)::AMDGPUPrintfRuntimeBinding::lowerPrintfForGpu(llvm::Module&) /b/sanitizer-x86_64-linux-fast/build/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp:576:39
#1 0x12f8287 in (anonymous namespace)::AMDGPUPrintfRuntimeBinding::runOnModule(llvm::Module&) /b/sanitizer-x86_64-linux-fast/build/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp:613:10
#2 0x544a96d in runOnModule /b/sanitizer-x86_64-linux-fast/build/llvm/lib/IR/LegacyPassManager.cpp:1750:27
#3 0x544a96d in llvm::legacy::PassManagerImpl::run(llvm::Module&) /b/sanitizer-x86_64-linux-fast/build/llvm/lib/IR/LegacyPassManager.cpp:1863
#4 0x9e58b7 in main /b/sanitizer-x86_64-linux-fast/build/llvm/tools/opt/opt.cpp:892:12
#5 0x7f84fa91f2e0 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x202e0)
#6 0x91ce99 in _start (/b/sanitizer-x86_64-linux-fast/build/llvm_build_msan/bin/opt+0x91ce99)

http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/34295/steps/check-llvm%20msan/logs/stdio

Thanks! I will check it.

Mon, Aug 12, 5:31 PM · Restricted Project
rampitec added a comment to D24035: [AMDGPU] Printf runtime binding pass.

Hi, MSan is complaining about this change:
==28574==WARNING: MemorySanitizer: use-of-uninitialized-value

#0 0x130075d in (anonymous namespace)::AMDGPUPrintfRuntimeBinding::lowerPrintfForGpu(llvm::Module&) /b/sanitizer-x86_64-linux-fast/build/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp:576:39
#1 0x12f8287 in (anonymous namespace)::AMDGPUPrintfRuntimeBinding::runOnModule(llvm::Module&) /b/sanitizer-x86_64-linux-fast/build/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp:613:10
#2 0x544a96d in runOnModule /b/sanitizer-x86_64-linux-fast/build/llvm/lib/IR/LegacyPassManager.cpp:1750:27
#3 0x544a96d in llvm::legacy::PassManagerImpl::run(llvm::Module&) /b/sanitizer-x86_64-linux-fast/build/llvm/lib/IR/LegacyPassManager.cpp:1863
#4 0x9e58b7 in main /b/sanitizer-x86_64-linux-fast/build/llvm/tools/opt/opt.cpp:892:12
#5 0x7f84fa91f2e0 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x202e0)
#6 0x91ce99 in _start (/b/sanitizer-x86_64-linux-fast/build/llvm_build_msan/bin/opt+0x91ce99)

http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/34295/steps/check-llvm%20msan/logs/stdio

Mon, Aug 12, 5:09 PM · Restricted Project
rampitec committed rG5b32752d107a: [AMDGPU] removed unused functions from printf lowering (authored by rampitec).
[AMDGPU] removed unused functions from printf lowering
Mon, Aug 12, 4:33 PM
rampitec committed rL368633: [AMDGPU] removed unused functions from printf lowering.
[AMDGPU] removed unused functions from printf lowering
Mon, Aug 12, 4:32 PM
rampitec closed D66117: [AMDGPU] removed unused functions from printf lowering.
Mon, Aug 12, 4:32 PM · Restricted Project
rampitec created D66117: [AMDGPU] removed unused functions from printf lowering.
Mon, Aug 12, 3:53 PM · Restricted Project
rampitec committed rGef8f1c473a82: [AMDGPU] Use PredicateControl in MIMGBaseOpcode. NFC. (authored by rampitec).
[AMDGPU] Use PredicateControl in MIMGBaseOpcode. NFC.
Mon, Aug 12, 3:34 PM
rampitec committed rL368626: [AMDGPU] Use PredicateControl in MIMGBaseOpcode. NFC..
[AMDGPU] Use PredicateControl in MIMGBaseOpcode. NFC.
Mon, Aug 12, 3:31 PM
rampitec closed D66115: [AMDGPU] Use PredicateControl in MIMGBaseOpcode. NFC..
Mon, Aug 12, 3:31 PM · Restricted Project
rampitec created D66115: [AMDGPU] Use PredicateControl in MIMGBaseOpcode. NFC..
Mon, Aug 12, 3:19 PM · Restricted Project
rampitec committed rG4c9c98f36b64: [AMDGPU] Printf runtime binding pass (authored by rampitec).
[AMDGPU] Printf runtime binding pass
Mon, Aug 12, 10:14 AM
rampitec committed rL368592: [AMDGPU] Printf runtime binding pass.
[AMDGPU] Printf runtime binding pass
Mon, Aug 12, 10:13 AM
rampitec closed D24035: [AMDGPU] Printf runtime binding pass.
Mon, Aug 12, 10:13 AM · Restricted Project

Thu, Aug 8

rampitec updated the diff for D24035: [AMDGPU] Printf runtime binding pass.

Cleaned, updated and rebased.
Pass does not require prelink infrastructure anymore.

Thu, Aug 8, 11:01 AM · Restricted Project
rampitec commandeered D24035: [AMDGPU] Printf runtime binding pass.
Thu, Aug 8, 10:58 AM · Restricted Project

Wed, Aug 7

rampitec accepted D65901: AMDGPU/SILoadStoreOptimizer: Add const to more functions.

LGTM

Wed, Aug 7, 1:37 PM · Restricted Project

Fri, Aug 2

rampitec accepted D65620: [AMDGPU] Use S_DENORM_MODE for gfx10.

LGTM

Fri, Aug 2, 11:47 AM · Restricted Project
rampitec added inline comments to D65600: Relax load store vectorizer pointer strip checks.
Fri, Aug 2, 9:49 AM · Restricted Project

Thu, Aug 1

rampitec committed rG6fe00a21f243: Handle casts changing pointer size in the vectorizer (authored by rampitec).
Handle casts changing pointer size in the vectorizer
Thu, Aug 1, 9:06 PM
rampitec committed rL367646: Handle casts changing pointer size in the vectorizer.
Handle casts changing pointer size in the vectorizer
Thu, Aug 1, 9:03 PM
rampitec closed D65612: Handle casts changing pointer size in the vectorizer.
Thu, Aug 1, 9:03 PM · Restricted Project
rampitec added inline comments to D65620: [AMDGPU] Use S_DENORM_MODE for gfx10.
Thu, Aug 1, 5:37 PM · Restricted Project
rampitec added inline comments to D65620: [AMDGPU] Use S_DENORM_MODE for gfx10.
Thu, Aug 1, 4:58 PM · Restricted Project
rampitec updated the diff for D65612: Handle casts changing pointer size in the vectorizer.

Changes to sextOrTrunc().

Thu, Aug 1, 4:43 PM · Restricted Project
rampitec added a comment to D65612: Handle casts changing pointer size in the vectorizer.

This one will go through our internal extended testing first, just to be sure it did not break anything.

Thu, Aug 1, 3:31 PM · Restricted Project
rampitec added inline comments to D65600: Relax load store vectorizer pointer strip checks.
Thu, Aug 1, 3:27 PM · Restricted Project
rampitec created D65612: Handle casts changing pointer size in the vectorizer.
Thu, Aug 1, 3:27 PM · Restricted Project
rampitec committed rGeee9312a85c0: Relax load store vectorizer pointer strip checks (authored by rampitec).
Relax load store vectorizer pointer strip checks
Thu, Aug 1, 3:19 PM
rampitec committed rL367624: Relax load store vectorizer pointer strip checks.
Relax load store vectorizer pointer strip checks
Thu, Aug 1, 3:19 PM
rampitec closed D65600: Relax load store vectorizer pointer strip checks.
Thu, Aug 1, 3:19 PM · Restricted Project
rampitec added inline comments to D65600: Relax load store vectorizer pointer strip checks.
Thu, Aug 1, 3:04 PM · Restricted Project
rampitec added inline comments to D65600: Relax load store vectorizer pointer strip checks.
Thu, Aug 1, 2:35 PM · Restricted Project
rampitec added inline comments to D65600: Relax load store vectorizer pointer strip checks.
Thu, Aug 1, 1:56 PM · Restricted Project
rampitec updated the diff for D65600: Relax load store vectorizer pointer strip checks.

Test update.

Thu, Aug 1, 1:56 PM · Restricted Project
rampitec added inline comments to D65600: Relax load store vectorizer pointer strip checks.
Thu, Aug 1, 12:56 PM · Restricted Project
rampitec created D65600: Relax load store vectorizer pointer strip checks.
Thu, Aug 1, 12:17 PM · Restricted Project

Wed, Jul 31

rampitec added a reviewer for D65496: AMDGPU/SILoadStoreOptimizer: Add helper functions for working with CombineInfo: vpykhtin.
Wed, Jul 31, 3:00 PM · Restricted Project
rampitec committed rGba1e845c2118: [AMDGPU] Fix for vectorizer crash with pointers of different size (authored by rampitec).
[AMDGPU] Fix for vectorizer crash with pointers of different size
Wed, Jul 31, 9:36 AM
rampitec committed rL367443: [AMDGPU] Fix for vectorizer crash with pointers of different size.
[AMDGPU] Fix for vectorizer crash with pointers of different size
Wed, Jul 31, 9:33 AM
rampitec closed D65480: [AMDGPU] Fix for vectorizer crash with pointers of different size.
Wed, Jul 31, 9:33 AM · Restricted Project

Tue, Jul 30

rampitec committed rG2594fa85932a: [AMDGPU] Fix high occupancy calculation and print it (authored by rampitec).
[AMDGPU] Fix high occupancy calculation and print it
Tue, Jul 30, 6:10 PM
rampitec committed rL367381: [AMDGPU] Fix high occupancy calculation and print it.
[AMDGPU] Fix high occupancy calculation and print it
Tue, Jul 30, 6:06 PM
rampitec closed D65423: [AMDGPU] Fix high occupancy calculation and print it.
Tue, Jul 30, 6:06 PM · Restricted Project
rampitec updated the diff for D65480: [AMDGPU] Fix for vectorizer crash with pointers of different size.
Tue, Jul 30, 5:50 PM · Restricted Project
rampitec added inline comments to D65480: [AMDGPU] Fix for vectorizer crash with pointers of different size.
Tue, Jul 30, 5:50 PM · Restricted Project
rampitec added a comment to D65480: [AMDGPU] Fix for vectorizer crash with pointers of different size.

Was this broken by r365723? Should this use stripPointerCastsSameRepresentation instead?

Tue, Jul 30, 3:38 PM · Restricted Project
rampitec created D65480: [AMDGPU] Fix for vectorizer crash with pointers of different size.
Tue, Jul 30, 3:30 PM · Restricted Project
rampitec committed rG9aff33bb9503: [AMDGPU] Print register pressure for agpr and vgpr separately (authored by rampitec).
[AMDGPU] Print register pressure for agpr and vgpr separately
Tue, Jul 30, 1:46 PM
rampitec committed rL367355: [AMDGPU] Print register pressure for agpr and vgpr separately.
[AMDGPU] Print register pressure for agpr and vgpr separately
Tue, Jul 30, 1:44 PM
rampitec closed D65476: [AMDGPU] Print register pressure for agpr and vgpr separately.
Tue, Jul 30, 1:44 PM · Restricted Project
rampitec created D65476: [AMDGPU] Print register pressure for agpr and vgpr separately.
Tue, Jul 30, 1:27 PM · Restricted Project
rampitec committed rG450afcea39ed: [AMDGPU] Reserve all AGPRs on targets which do not have them (authored by rampitec).
[AMDGPU] Reserve all AGPRs on targets which do not have them
Tue, Jul 30, 12:30 PM
rampitec committed rL367347: [AMDGPU] Reserve all AGPRs on targets which do not have them.
[AMDGPU] Reserve all AGPRs on targets which do not have them
Tue, Jul 30, 12:30 PM
rampitec closed D65471: [AMDGPU] Reserve all AGPRs on targets which do not have them.
Tue, Jul 30, 12:30 PM · Restricted Project
rampitec added reviewers for D65423: [AMDGPU] Fix high occupancy calculation and print it: mareko, dstuttard.
Tue, Jul 30, 12:19 PM · Restricted Project
rampitec created D65471: [AMDGPU] Reserve all AGPRs on targets which do not have them.
Tue, Jul 30, 12:18 PM · Restricted Project
rampitec accepted D65467: AMDGPU: Remove v0 workaround for DS_GWS_* instructions.

LGTM

Tue, Jul 30, 12:07 PM
rampitec accepted D65454: AMDGPU: Add missing builtin declarations.

LGTM

Tue, Jul 30, 9:30 AM

Mon, Jul 29

rampitec updated the diff for D65423: [AMDGPU] Fix high occupancy calculation and print it.

Removed "GCNSubtarget::getMaxWavesPerEU()" in favor of using "AMDGPUSubtarget::getMaxWavesPerEU;"

Mon, Jul 29, 5:33 PM · Restricted Project
rampitec updated the diff for D65423: [AMDGPU] Fix high occupancy calculation and print it.

Added field MaxWavesPerEU instead of virtualizing the function.

Mon, Jul 29, 5:31 PM · Restricted Project
rampitec added inline comments to D65423: [AMDGPU] Fix high occupancy calculation and print it.
Mon, Jul 29, 5:24 PM · Restricted Project
rampitec added inline comments to D65423: [AMDGPU] Fix high occupancy calculation and print it.
Mon, Jul 29, 4:30 PM · Restricted Project
rampitec created D65423: [AMDGPU] Fix high occupancy calculation and print it.
Mon, Jul 29, 3:51 PM · Restricted Project
rampitec added inline comments to D65170: Migrate some more fadd and fsub cases away from UnsafeFPMath control to utilize NoSignedZerosFPMath options control.
Mon, Jul 29, 12:53 PM · Restricted Project